Internet Shopping Optimization Project (IShOP)

Jedrzej Musial

, Johnatan Pecero

, Bernabe Dorronsoro

and Jacek Blazewicz

Institute of Computing Science, Poznan University of Technology,

ul. Piotrowo 2, 60-965 Poznan, Poland

Computer Science and Communications Research Unit, University of Luxembourg,

6 rue Coudenhove-Kalergi, L-1359 Luxembourg, Luxembourg

Jedrzej.Musial@cs.put.poznan.pl

Abstract. E-commerce (or even e-business) is becoming a part of modern soci-

ety. The continuously growing implementation of technology (e.g., cloud com-

puting, mobile devices - smartphones, tablets) into our daily business and admin-

istrative operations make it necessary to adapt to this inevitable evolution. One

of its essential parts is Internet shopping, which becomes more and more popu-

lar with every upcoming year. Access to the service industry is now also offered

through internet portals, ranging from cloud computing to translation services.

Shipping costs, quantity discounts and early booking, among others, allowed for

the creation of value added-services based on brokering. This project proposes

innovative and realistic models for different typical online shopping operations,

supported by strong mathematical and operational research fundamentals, and

well balanced with lightweight computational algorithms. These models are de-

signed in order to allow the optimization of such transactions. Finding accurate

solutions to the deﬁned problems implies both lowering customer expenses and

favouring market competitiveness. Therefore, the outcome of the project will be

extremely beneﬁcial for the society; particularly taking into account that online

shopping already comprises a large percentage of the actual commerce (in 2013,

50% of European consumers will be making purchases online). One of the main

aims of this project is to model and formulate new advanced and realistic ﬂavours

of the Internet Shopping Optimization Problem (ISOP), considering discounts

and additional conditions like price sensitive shipping costs, incomplete offers

from shops, or the minimization of the total realization time, price, and deliv-

ery time functions, among others. The models will be mathematically and the-

oretically well founded. Moreover, the challenge of deﬁning and addressing a

multi-criteria version of the problem will be addressed too. Other important con-

tributions will be the mapping of ISOP to other new challenges. One of them is

the design of a novel business model for cloud brokering that will beneﬁt both

cloud providers and consumers. Providers will be able to easily offer their large

number of services, and to get a fast answer from the market to offers (e.g., when

infrastructure is under-utilized). Additionally, customers will easily beneﬁt from

offers and ﬁnd the most appropriate deals for his/her needs (according to ser-

vice level agreements, pricing, performance, etc.). Modelling some of these as-

pects and coupling it with an optimization tool for the brokering of cloud services

among various providers would be a key contribution to the ﬁeld. Finally, a wide

set of optimization algorithms will be designed and developed for the addressed

problems. They include from fast lightweight specialized heuristics to highly ac-

curate parallel and multi-objective population-based metaheuristics. They all will

Dorronsoro B., Pecero J., Musial J. and Blazewicz J.

Internet Shopping Optimization Project (IShOP).

DOI: 10.5220/0006144400160033

In European IST Projects - The Quest for Excellence Towards 2020 (EPS Vienna 2014), pages 16-33

ISBN: 978-989-758-101-4

be embedded in a software framework for their practical applications, and vali-

dation.

1 Accordance with the Project Call - PolLux

This project deals with the e-services sector (both public and private), addressing the

important market of online shopping, by modeling and optimizing very common re-

lated problems. The studies of the project are supported by a strong mathematical back-

ground, both in the deﬁnition and study of the problems, as well as in the demonstration

of their NP-hardness. Additionally, efﬁcient resolutions of such complex problems re-

quire the development of different novel IT solutions related to the effective exploration

of huge search spaces. A number of both fast and highly accurate smart methods to ﬁnd

near-optimal solutions to the problems will be investigated. The considered topic com-

pletely matches the business service design thematic research priority of the current

POLLUX call. The solution we will investigate for the internet shopping optimization

problem will be a business e-service of great beneﬁt for online customers. They will

enhance competitiveness among online shops. Additionally, we propose a novel busi-

ness model for cloud brokering. The cloud broker will provide his customers with the

best option to deploy their services or execute their algorithms in the cloud. The choice

will be made considering not only cost, but also quality of service, reliability, security,

etc. Moreover, the broker will negotiate with the cloud providers unusually low prices

thanks to the large market volume he generates. Cloud providers might also be highly

interested in dealing with the broker special offers to get customers in some cases (e.g.,

when resources are being under-utilized). The project aims to design a novel business

model for cloud brokering, which supports the Innovation in Services Research Priority.

The engineered solutions are valid for both public and private e-services.

2 Description of the Project

2.1 Current State of the Art Including Your Relevant Previous Work

Introduction: Electronic Commerce

One can say that electronic commerce (e-commerce) is one of the fastest developing

ﬁelds of computing science (based on operational research, combinatorial optimiza-

tion). However, it is worth noticing that e-commerce is a giant hybrid built in additional

areas such as logistics, economy and social sciences. E-commerce is an industry, which

focuses on selling and buying products and services through web pages [1, 2]. Online

shopping, ﬁtting into a business-to-consumer (B2C) sub-category, is one of the key

business activities offered over the Internet. It has become increasingly popular over

the past decade. U.S. e-commerce market sales amounted to 289 billion USD in 2012,

up from 256 billion USD in 2011. In Poland, Internet retailing in goods amounted to

2 billion Euros in 2010. Luxembourg is a strategic place for e-commerce and online

shopping with the key players like Amazon, Skype, iTunes, PayPal, eBay, and many

other leading ICT companies [3]. This attractiveness is mainly due to the state-of-the-

art infrastructure, telecommunication network (e.g. TeraLink network), one of the best

Internet Shopping Optimization Project (IShOP)

data center parks in EU to support cloud computing, trust culture (with LuxTrust Lux-

embourg being the ﬁrst European country to provide online certiﬁcation for the private

sector) and one of the lowest VATs in the EU.

Internet Shopping Optimization

Price comparison sites (e.g., Google Shopping, Shopping.com, PriceGrabber, or Kelkoo

are among the most popular shopping engines in EU) are search tools designed to give

price information from many retailers through a single portal. However, current price

ranking solutions target only single product buying. A signiﬁcant percentage of price

comparison websites perform sub-optimally in one of their major functions: presenting

prices and considering multiple products.

Motivated by the problem of buying multiple products from different e-commerce

web sites, we proposed a combinatorial optimization problem where we have a set of

products N = 1, . . . , n and a consumer who wishes to buy all products in N from a set

of Internet stores M = 1, . . . , m at the minimum ﬁnal price, subject to the price, avail-

ability, and discount constraints at each store [4–6] (For a double discounting function

one can follow [7]). For i = 1, . . . , m, let S

denotes the decision variable that corre-

sponds to the subset of products selected by the consumer from store i. The correspond-

ing optimization problem, which the authors refers to as the ISOP (Internet Shopping

Optimization Problem), is given by:

min

i=1

j∈N

s.t.

i∈M

= 1, j = 1, . . . , n,

0 ≤ x

≤ y

, i = 1, . . . , m, j = 1, . . . , n,

∈ {0, 1}, y

∈ {0, 1}, i = 1, . . . , m, j = 1, . . . , n.

where d

denotes the delivery price from store i and N

⊆ N denotes the set of

products available at store i. Moreover, for each i and j, p

denotes the price of product

j in store i. The function f

: R

→ R

denotes the discount function associated with

store i (which is a function of the delivery and total price of the products).

During previous research different versions (specializations) of ISOP were exam-

ined. All of them are described in journal publications as well as presented in a number

of international conferences [8, 9]. The main stress was placed on the most complicated

version of the problem (the most realistic at the same time) where we deal with the

discounting functions for every store.

It is worth noticing that there are some similarities with the well-known Facility

Location Problem (FLP) [10]. The main characteristics of the FLP are space, the met-

ric, given customer locations and given or not given positions for facility locations.

A traditional FLP is to open a number of facilities in arbitrary positions of the space

(continuous problem) or in a subset of given positions (discrete problem) and to assign

customers to the opened facilities so that the sum of opening costs and costs related to

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

the distances between customer locations and their corresponding facility locations is

minimized.

Discussions of FLPs can be found in [11, 12, 10, 13, 14]. The traditional discrete

FLP is NP-hard [15, 16] in the strong sense. It was also proved that ISOP without any

discounts and ﬂat shipping rates is NP-hard as well [4]. Note, however, that the general

problem Basic ISOP with price sensitive discounts cannot be treated as a traditional

discrete FLP because there is no evident motivation for a discount on the cumulative

cost in the sense of distances. It can be noticed that this problem and problem Basic

ISOP are not sub-cases of one another, while the traditional discrete FLP is a special

case of any of these problems.

Internet shopping problems were under interest of the working group already for

some time. As a result of this work one can mention several journal publications [4–

6]. Moreover, Internet shopping problems, mathematical formulations, complexity dis-

cussions, as well as algorithms propositions and some ﬁrst computational tests were

done within a Ph.D. thesis by an applicant entitled “Applications of Combinatorial Op-

timization for Online Shopping” [17]. What is surely worth noticing is that the the-

sis was written in a co-tutelle programme between Poznan University of Technology

and the University of Luxembourg. Furthermore, good collaboration between both uni-

versities groups brought about another publication [5, 6]. Results of the collaboration

have already been presented during some OR / combinatorial optimization conferences

(ECCO 2012, EURO 2012) and this trend will certainly be continued (already accepted

for presentations this year during ECCO 2013, MISTA 2013, OR 2013). The main con-

tributions of the previous work within Internet shopping problems are the following:

– Mathematical formulation of the Internet Shopping Optimization Problem (ISOP),

– providing the proof that the ISOP problem is NP-hard by a polynomial transforma-

tion from the Exact Cover by 3-Sets (X3C), which is a well-known NP-complete

problem,

– detailed literature analysis with a special attention paid to Facility Location and

Knapsack Problems,

– algorithm design a preliminary version of the algorithm solving basic ISOP was

prepared.

Cloud and Cloud Brokering

Cloud computing [18, 19] is, undoubtedly, one of the main existing computing paradigms

nowadays. In the last years, it raised the interest of both academic and industrial worlds

thanks to their interesting properties, such as elasticity, ﬂexibility, or computational

power, among many others. Cloud computing provides a stack composed of different

kinds of services to users [20]: Infrastructure as a Service (IaaS), dealing with resources

as servers, storage, or networks; Platform as a Service (PaaS), which provide an oper-

ating system as well as a set of tools and services to the user; or Software as a Service

(SaaS), that allows providers to grant customers with access to licensed software.

Many different public and private clouds are arising in the last years [21]. They all

have distinct features, making it difﬁcult for users to ﬁnd the best choice among all the

existing services offered by the Cloud Service Providers (CSPs) [22]. The reason is that

cloud users must deﬁne their speciﬁc requirements in terms of services and application

Internet Shopping Optimization Project (IShOP)

deployment [23]. These speciﬁc requirements can be categorized into measurable fea-

tures such as cost or resources and user-experience features such as response time or

performance (based on the feedback from the cloud users experience about the service

they purchased from the CSPs). These parameters must be collected from the data log of

the VMs/Services of the cloud user on-running systems using machine learning system

to compare with promised QoS of the CSPs.

An example of the measurable features is presented in the study of Chaisiri et al.

[24]. The authors pointed out that the pricing model offered by CSPs includes on-

demand and reserved prices. However, cloud users also need to take into considera-

tion the on-running cost for the instances that they purchased from CSPs. So there are

three phases for the life-cycle of VM management in CC: reservation, on-running and

on-demand phase [25, 26]. Based on this model, the authors presented the differences

between private and public CC not only in the price plan but also in the resource limi-

tation.

In cloud computing, CSPs must offer their services to cloud users with the promised

SLA and Quality of Service (QoS). There are a few differences in cost or plan between

CSPs [27]. The characteristics of the services offered by CSPs can be deﬁned as feature

and non-feature characteristics. The feature characteristics are price model (average

monthly price), resource plan and SLA/QoS. The non-feature characteristics can be

deﬁned as virtualization, performance, reliability and security. CSPs can also be cate-

gorized with respect to the service model: SaaS, PaaS and IaaS, or the QoS and SLA

for the services they provide.

The ﬁgure of cloud broker [28] arises as an intermediary entity between cloud

providers and users to help the latter ones in the process of choosing the most appro-

priate services among those offered by the different CSPs, according to their particular

needs. There are different services that cloud brokers can provide, from simply ﬁnding

the best deals among a set of clouds for the user requirements to deﬁning the best pos-

sible design to deploy the user’s application in the cloud [28]. Additionally, the cloud

broker can consider the QoS and SLA from the CSPs as parameters to compare with

the results from the historical data or log analysis based on the on-running services of

the cloud users, in order to ensure high QoS.

There are numerous available studies in the ﬁeld of research for the cloud broker. In

the research conducted by Spillner et al. [29], the authors show that the cloud broker is

not only the interface to manage the virtualized resources between the cloud provider

and the cloud user but also can help to bring the unused resources from the cloud users to

reuse in the cloud. Therefore, they proposed to deﬁne a nested VM in which other VMs

from multi-cloud providers are referred to as sub-VMs. The concept of nested VMs

helps the cloud broker to deal with a variety of VMs from multi-cloud providers. In

these sense, there is a business model in which the cloud broker buys reserved instances

from a number of clouds and then sublet them as an on-demand basis to the users at

cheaper prices than those of the cloud providers. This is proﬁtable thanks to the high

price difference between on-demand and reserved VMs [30, 31].

Usha et al. [32] also proposed a framework for cloud brokerage service. The frame-

work schedules the cloud resources by considering both the multi-criteria objectives

from cloud users and cloud providers. Their proposed model based on the QoS parame-

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

ters includes: response time and the throughput. The optimization problem is deﬁned as

the multi-criteria optimization so that the author proposed to use the Pareto technique

(the Pareto front) to ﬁnd the optimal solutions.

Carpentier et al. [33] presented the CompatibleOne, an open-source framework for

the cloud brokerage service to work with a multi-cloud environment. Two important

components were proposed in the CompatibleOne: (a) the energy-monitoring module

and (b) the module to interact with the cloud monitoring systems.

Cloud brokers are intermediary entities between the cloud providers and their cus-

tomers that provide different services to the users of the cloud. From the previous lit-

erature review we can notice that different service management problems have been

considered from the cloud brokering point of view. We deﬁne in this project a ser-

vice management (cloud brokering) problem by considering a customer requiring a set

of (different) services offered on the cloud by a cloud provider or by a set of private

and public cloud providers with an associated service cost (different for each service

provider). The aim is to minimize the service cost from the customer point of view

and maximize resource utilization from the service provider point of view. The service

management problem can be adopted by cloud brokers to provide a better service to

the customers and deal low prices with cloud providers, who at the same time might

beneﬁt from such a broker negotiating with customer demands in case of under- or

over-utilization of resources. The simpliﬁed cloud brokering problem can be reduced to

the ISOP problem such that, algorithms and approaches developed to solve ISOP can

be applied to the cloud brokering problem. However, advanced models of cloud bro-

kering need to consider carefully different service level agreements between customers

and service providers, as well as to deal with different quality of services (services re-

quirements, resources availability, discounts, etc.) that will affect the optimization of

the problem.

2.2 Project Objectives and Contribution to Knowledge Development

The project is expected to contribute to the existing knowledge in different ﬁelds, as it

is outlined in Figure 1.

From the application point of view, we expect outstanding contributions related to

the problem of providing optimal planning for internet shopping (the ISOP problem).

This is an important problem for society, especially nowadays that internet shopping

is becoming more and more popular and convenient, that has not yet been adequately

addressed in the scientiﬁc literature. This project will focus on the modeling of the

problem, with very strong mathematical foundations, on the study of its main features,

and the development of novel highly efﬁcient heuristics to provide accurate solutions.

Therefore, important contributions are expected in the ﬁeld of operational research. One

important objective of this project is to provide other researchers with a solid highly re-

alistic model for this important problem. We expect many different research groups

worldwide will be interested in the research line we plan to open thanks to this project.

Several new subproblems of the ISOP will be deﬁned depending on particular parame-

ters. To be more precise, we plan to:

– Analyze requirements and describe extended models for the ISOP problem.

Internet Shopping Optimization Project (IShOP)

Fig. 1. Summary of the main contributions of IShOP project.

– Formulate Internet Shopping Optimization Problem with discounts and with addi-

tional conditions:

• Price sensitive shipping cost idea that follows discounting function a few

thresholds with different shipping costs,

• Minimum delivery times,

• Incomplete shopping lists realization.

– Provide the models with mathematical and theoretical foundation.

– Prove NP-hardness.

– Analyze decision-aid multi-criteria ISOP. ISOP will be enhanced with total realiza-

tion time minimization of the total function (utility function):

• price and delivery times functions,

• decision-aid recommender function. Only higher recommendation rank shops

inﬂuence of the recommendation rank on the total price.

For these problems new algorithms will be prepared and tested. The results will be

presented in several journal publications. A software platform aimed at testing the new

approaches will be also developed.

As an attempt to disseminate the modeled ISOP problem in different ﬁelds and to

show its generality, we plan to map it into the brokering problem for cloud computing,

in order to help the broker to manage the huge existing offer of services. We are par-

ticularly interested in ﬁnding the best offers from a number of cloud providers that suit

the requirements of the customers. Moreover, the modeled cloud broker will provide

additional advantages, such as negotiating lower prices with cloud providers thanks to

the large market volume they generate. Cloud providers will also beneﬁt from the bro-

ker, since they will be able to offer the broker good terns to get customers in case of

underutilization of the resources, or to direct customers to other providers in peak times

when the resources are overutilized. This will be done in a transparent way to the cus-

tomer, so the reputation of the provider will not be damaged. The novel cloud services

management agent we will work on is applicable to any cloud broker in order to make

it much more advanced that the ones existing in the literature, and therefore it will be

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

a groundbreaking result. The scientiﬁc literature about cloud brokering is large, and it

is one of the hot topics in the cloud computing ﬁeld, because it is considered as one of

the mechanisms that will allow the federation and interoperability of clouds in the fu-

ture. Additionally, cloud brokering is a novel business model, and many new companies

appeared in the market in the last years.

The proposed optimization for cloud broker will be based on the development of the

ISOP, and it should deal with conﬂicting objectives that will beneﬁt both cloud users

and CSPs. This is therefore a multi-objective optimization problem that will be solved

with simple heuristics [34], sequential and parallel local search algorithms [35, 36], and

Evolutionary Algorithms [37–41]. The algorithms will look for the optimal Pareto sets

of solutions that satisfy both cloud users and CSPs.

The optimization problem of the cloud users and the CSPs includes both the measur-

able and user-experience parameters. The measurable parameters are cost and resource

placement. While the user-experience parameters contain response time, performance

and security. The term user-experience means that these parameters will be based on

the user experience about the services they purchased from the CSPs, these parame-

ters will be included into a machine learning model with scoring function about the

provided services of CSPs. The proposed mathematical model is studied to deal with

the optimization problem that contains measurable parameters: cost and virtual ma-

chine (VM) placement. Moreover, a machine learning technique will be used to deal

with user-experience parameters from the results of historical data or log based on the

on-running services of cloud users.

The concept of the proposed cloud brokering methods can be deﬁned by character-

istics as:

– Interface: the cloud broker is the interface between cloud users and CSPs,

– the cloud broker plays an important role to negotiate in order to balance the opti-

mization objectives of cloud users and CSPs. It is obvious that there are conﬂicts

between both parties objectives. Therefore the cloud broker offers a multi-objective

optimization method to ﬁnd optimal tradeoff solutions from the Pareto front,

– choosing and assigning tasks/services from the requests of cloud users to deploy on

the services of CSPs,

– managing VM/Service repository with service classiﬁcation (the VMs or services

run the same task),

– managing cloud services stack from the available CSPs.

In the design concept of the cloud broker, there is a VM/Service repository that

stores VM or service speciﬁc requirements and cloud service stack stores details of

available CSPs. The VM/Service repository stores parameters including:

– Measurable parameters:

• Resource vector v = CPU, RAM, storage, network inbound/outbound data

• Cost: pricing model reservation, on-demand, on-running

• Service model: SaaS, PaaS, IaaS

– User-experience parameters:

• Security

• Plan (Small/Medium/High/etc.)

Internet Shopping Optimization Project (IShOP)

• User-location

• Time peak/low schedule

In order to successfully carry out our investigations for this project, we will need

to make other important contributions in a number of different ﬁelds as decision aid,

multi-criteria decision making, multi-objective optimization or parallel optimization al-

gorithms, among others.

2.3 Methods and Approach

Problems to be considered in this proposal, involve many constraints and objective func-

tions, which are deﬁned on sets of discrete objects rather than on continuous domains.

Problems of this kind have a combinatorial nature. Very often problems of this nature

are hard to solve because they require verifying a number of possible solutions, which

grow exponentially with the problem size. Since exponential functions grow almost

explosively, it is practically impossible to solve such problems by full enumeration of

possible solutions. Such approaches are called exact. Unfortunately, they are practically

ineffective. On the other hand, effective algorithms are the ones in which execution time

can be bounded by a polynomial in the input size. Unfortunately, according to the cur-

rent state of knowledge, there exist hard combinatorial problems for which no effective,

or in other words polynomial, algorithms exist.

The theory of computational complexity provides methodology of dealing with hard

combinatorial problems. In particular, the methods of recognizing hard combinatorial

problems, discerning their different classes and providing types of usable algorithms

for certain classes. The ﬁrst step in analyzing a combinatorial problem is to verify if it

is solvable in polynomial time. If not, then a proof should be presented that the problem

belongs to a class of hard problems such as NP-complete or NP-hard problems. In such

a case the problem can be dealt with in a few ways: an exponential running time may

be proposed, however with a limited applicability. A heuristic may be also proposed.

Heuristics are a group of algorithms that run in polynomial time, and hence they are

effective and provide correct solutions, yet without a guarantee of optimality. Heuristic

algorithms can be analyzed with the goal of providing guarantees of solution quality,

e.g. that the ratio of the value of the solution provided by the heuristic and the optimum

objective value is bounded.

Problems considered in this project are also hard combinatorial optimization prob-

lems. Therefore, they will be analyzed according to the above methodology. First, the

complexity class of the particular problem must be determined. In a best case the prob-

lem may be solvable by an effective (polynomial) algorithm. Yet, due to the nature of

the solved problems and their computational hardness it is more likely that other algo-

rithms will be applied: exact, exponential-time or heuristic, polynomial-time methods.

Efﬁcient solving of practical problems, such as internet shopping and cloud broker-

ing require, ﬁrst of all, precise and concise formulation. The systematic research will

be preceded by proposing mathematical models of considered problems, which reﬂect

the real world environment in the set of variables and constraints. The general models

allow collecting the most important components and features describing the problem

and they are the basis for some relaxations and direct the future research. Depending

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

on the ranges of values, their distributions and the impact of particular constraints on

some variants of the problems might be distinguished, varying in their complexity. It

would be interesting to investigate computationally easy cases, which polynomial time

algorithms, might be used as sub-procedures allowing for faster solving of more gen-

eral cases. Moreover, the analysis of the mathematical models should support searching

for similarities between the considered problems and closely related optimization prob-

lems, such as the travelling salesman or purchaser problems, in order to incorporate

some methods developed for these cases. Since real world problems are usually compu-

tationally hard, the research will focus on constructing heuristic (efﬁcient in quality) and

exact (efﬁcient in time) algorithms, after formal proving NP-hardness of the analyzed

cases. For algorithms used, e.g., in on-line environment for internet shopping - the time

complexity is a crucial factor, since it is necessary to ensure acceptable response time.

For methods used e.g. for advertisement placement, the quality of solutions is more

important than their run times. Depending on the considered case, proper algorithmic

strategies will be used: fast simple heuristics or more sophisticated and consequently

slower metaheuristics, and even branch and bound methods. In some cases, efﬁcient

dynamic programming approaches will be taken into account, which should give op-

timal solutions in reasonable time. Mathematical programming formulations might be

also useful for incorporating the models into commercial solvers such as e.g. CPLEX,

which might serve as source of reference solutions.

The above discussion shows that three groups of algorithms may be applied to solve

hard combinatorial optimization problems:

1. Exact algorithms. The main feature of exact algorithms is that they search the space

of possible solutions exhaustively. It means that in the worst case all possible solu-

tions may be visited. Therefore, in the worst case, execution time of such algorithms

may be prohibitively high. However, there are techniques allowing for pruning the

search space. For example, in branch-and-bound algorithms the space of solutions

is represented as a search tree. The tree is pruned by using bounds on values of

solutions that may be derived from a given node of the tree. Branch-and-bound

algorithms were applied in the past with a relative success. Hence, wisely crafted

exact algorithm may be practically usable.

2. Dedicated heuristics are a group of methods, which are created to solve just one

combinatorial optimization problem. Due to different characteristic of the prob-

lems it is not possible to transfer such heuristic to a different problem. Still, by

their polynomial running time they are very often applied to solve combinatorial

optimization problems.

3. Metaheuristics are a group of general purpose methods which can be adapted to

many combinatorial optimization problems. Metaheuristics provide general schemes

of conducting search in the space of combinatorial problems. Since they are very

general, it is necessary to provide an interface to the actual combinatorial optimiza-

tion problem. Example metaheuristics are simulated annealing, tabu search, genetic

search, ant systems and many others. This type of algorithmic approach to combi-

natorial optimization was very successful and has been extensively studied in recent

years.

Internet Shopping Optimization Project (IShOP)

To tackle the complex NP-hard problems deﬁned in the project, advanced optimiza-

tion methods need to be developed. High effort will be made to design highly efﬁcient

and accurate heuristic algorithms to solve the considered problems. A guarantee of the

performance of such tools will be provided, for instance by validating the results they

provide with the optimal solution, computed by exact methods, when possible. The pos-

sibility of designing algorithms that guarantee a given maximum error on the solution

from the optimum will be explored too. Our previous experience on the design of opti-

mization algorithms for problems as tasks scheduling in high performance centers [39,

38, 42, 43, 23, 44, 22] or our preliminary ISOP problem [4–6, 8, 9, 17] will constitute an

asset.

The considered problems that we study in this project are multi-objective in nature.

This means that they usually require more than one objective to optimize, none more

important than the other, and that are usually in conﬂict (for a given optimal solution,

increasing the quality of one of the objectives leads to worsening some of the others).

We therefore plan to apply advanced novel multi-objective algorithms to ﬁnd a wide

and diverse range of accurate tradeoff solutions (i.e., a Pareto front approximation) in

order to study the shape of such set of solutions. This study is very useful because it

will allow us to get important knowledge on the problem such as knowing which kind

of tradeoff solutions we are more interested in (i.e. which region of the Pareto front

approximation), and then we can design efﬁcient heuristics to search that area. Addi-

tionally, parallel algorithms (both for multi-objective and single-objective optimization)

will be considered to speed-up the optimization process in the case they are required.

Moreover, the important part of the project (Software Framework) will be to im-

plement and experimentally validate a theory-driven online model in realistic testbeds

to demonstrate the performance improvement over currently used techniques. The task

is challenging and requires concentrate a large effort work with the algorithms design,

computational experiment, and data collecting engine preparation. The latest mentioned

task is a priority to perform the best possible quality of computational experiments.

These will conﬁrm the highest quality of proposed algorithms. Model load should be

perfectly balanced since it will be accessible online. During the mentioned package we

plan to use several approaches to achieve our goals:

– Develop efﬁcient lightweight heuristic based algorithms, considering approxima-

tion factors, greedy based algorithms.

– Investigate pseudo-parallel cellular based optimization algorithms (shorter compu-

tation time).

– Mathematical analysis of the performance of the algorithms.

– Computational tests of the developed algorithms.

– Research on smart algorithms to ﬁnd highly accurate solutions and target large

problem instances.

Performance of computer algorithms can be assessed in many different ways. Yet,

the most important aspects are related to the resources consumed by the algorithm and

to the quality of the results the algorithm provides. The resources may have different na-

ture: memory, processors, time. The quality of algorithm output may be also measured

in many ways: e.g. as the correctness of the output, frequency of errors, etc.

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

In the context of combinatorial optimization the key algorithm performance indi-

cators are runtime and distance of the solution from the optimum. Runtime, or the

algorithm execution time, is very important because exact algorithms, such as branch-

and-bound (B&B), require exponential time in the size of the input in the worst case.

However, successful B&B implementations are capable of mitigating this exponential

growth. Furthermore, it is a very common situation that a trade-off between solution

quality and the runtime exists, especially in the case of metaheuristics. Thus, some al-

gorithms may be more effective in trading time for the solution quality.

Algorithm execution time and solution quality distance from the optimum may be

studied in various ways. One possible approach is analytical. In this case, the order of

the computational complexity of the algorithm is searched for, as an indicator of the

algorithm temporal cost. The computational complexity of an algorithm is a function

of the algorithm execution time in the size of the input. The quality of the solutions

is assessed by analytically providing evidence (e.g. a proof) that the distance of the

solution quality from the optimum, in the worst or average case, is bounded.

Performance of algorithms may also be studied experimentally. In this case, the

algorithm is tested on benchmark instances. Its runtime and solution quality is compared

against other algorithms. Note that certain algorithms may be better on some test cases

and worse on others. Consequently, experimental analysis may be inconclusive when

the criteria of the comparison are not strictly deﬁned. Quite often statistical analysis of

the results is accepted as a method resolving such problems.

3 Description of the Project Plan and Work Packages

IShOP project is divided into 5 Work Packages, organized as it is shown in the Figure

Each of the Work Packages tackles a full topic. The tasks in the different Work

Packages are designed to be iterative and at a given level most of them can be done in

parallel. WP1 and WP5 are pure management tasks to focus on proper project guidance

and best possible presentation of the results (workshops, conferences, publications, re-

ports, websites, etc.).

We combine mathematical, theoretical research work/formulations/modeling in WP2

and WP3 with practical use of them (as preparation on test applications in Software

Framework WP4). Algorithm design and optimization are extremely important and we

will focus on them in WP4 as well as partially in the previous WPs. All WPs will pro-

vide results that will be important contributions to e-commerce and, more generally,

operational research, optimization and computer science. Objectives, deliverables and

milestones were deﬁned for the different WPs, and they are described in detail in each

of the Work Packages.

What is surely worth noticing is that the project will promote the collaboration be-

tween Poznan and Luxembourg into higher level of quality. Running IShOP project

standalone on each university would deﬁnitely not be feasible. The key factor was to

combine the best possible top-class specialists in each ﬁeld and make them WP leaders.

Moreover, the whole project team is carefully picked to complement their skills and to

lower the risk of failure to the minimum. Furthermore, each WP group is balanced with

Internet Shopping Optimization Project (IShOP)

Fig. 2. Project plan: packet interdependencies.

both Polish and Luxembourgish scientists in order to combine their expertise for suc-

cessfully accomplishing the objectives (each team represents different speciﬁc skills).

That will also allow to positively stimulate one another and provide new ideas. Ac-

cording to our experience, this approach is much better for the project quality and risk

management than working in hermetic groups. Some of the members involved in the

project from the two groups (UL and PUT) are already used to work together, and some

common publications proof that.

The Project PI is Prof. Dr. Jacek Blazewicz. His scientiﬁc experience (over 340 pa-

pers in outstanding journals, numerous international awards, and prestigious positions

in research societies, H-index=26) is a guarantee of a high chance for a successful out-

put of the grant. Prof. Dr. Pascal Bouvry (from UL) was the quality director of a major

software development company for CLES, R&D director and general manager for SDC

company, as well as PI for a number of research projects. His managerial and scientiﬁc

expertise will be key for IShOP.

4 Project Outputs

The ﬁrst contribution of this project will be a formulation of several new optimiza-

tion problems. All of these problems arise from practical concerns originating from

e-commerce, and they will be deeply analyzed from a practical point of view and well

deﬁned with the use of formal mathematical formulation. Therefore, we expect they

should open new interesting research areas. Proofs of NP-hardness will be provided.

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

For multicriteria problems or problems concerning optimization of several goals, dif-

ferent objective functions will be proposed. Whenever possible, approximation factors

will be considered. The above issues will offer a very good starting point for further

research, e.g. algorithmic, but also deeply practical.

The second output of this project will be the creation of the strong research group

conducting research in the area of algorithms and services for e-commerce and related

topics, such as cloud service optimization problems and website optimization problems,

originating from cooperating researchers from both institutions. Currently, these topics

are covered almost isolated at both universities, mostly by sole researchers. Still, results

achieved so far show that this area is interesting and important, only very underde-

veloped. The new situation would join experienced researchers, with wide theoretical

knowledge of algorithms and methods, with young ones (even students) with most up

to date understanding of e-commerce current trends and ideas, and thus allow to reach

the critical mass. This would also beneﬁt in researcher training for one PhD student and

seven postdocs, including some freshly ones after their dissertation, as well as highly

experienced ones. Both partners would greatly beneﬁt from this project in terms of de-

velopment and visibility. A non-disclosure agreement will be signed between the two

parties composing the consortium after the project is granted, in order to ensure a cor-

rect management of the intellectual property rights of the new achievements.

The foreseen publication output of this project is composed by at least three peer

reviewed research papers (related to WP2, WP3 and WP4) published in good qual-

ity journals from the ISI list and over 6 papers published in top conferences (peer re-

viewed). We strongly believe that research results, especially publicly funded, should

be as accessible as possible, thus we plan all the papers to be published in the open ac-

cess, when possible. Choice of journal with good impact factor will lead to subscription

model, fees for open access articles will be paid.

Both teams, PUT and UL, will have the opportunity to disseminate the knowledge

gained with the project to the classrooms (highly related subjects are taught by the re-

searchers involved in the project at PUT and LU, both at master and Ph.D. programs).

There will also be opportunities for some students to directly participate in the project

(student contracts are planned). Previous experience indicates that students show more

interest when knowledge is illustrated with practical applications, of course, with well-

founded theoretical background. Additionally, the results obtained from this project will

be of high interest for the two institutions to compliment programs already on-going at

both sites. The methodology developed and used in this project to analyze theoretical

performance guarantees and approximation factors for optimization algorithms is also

interesting for the goals of the research teams of the two institutions and valuable for

other research projects in ﬁelds related to security, bioinformatics, Grid and cloud com-

puting, in the frame of academic and industrial research projects (e.g., SUPER Node

II in partnership with MixVoIP company, a satellite payload optimization project in

partnership with SES satellite company, ECO-CLOUD FNR project, or INDECT FP7

project).

We expect key knowledge to be generated and developed in the framework of this

research work. The project will thus contribute to the advancement of state-of-the-art

knowledge production and its dissemination through peer reviewed publications, work-

Internet Shopping Optimization Project (IShOP)

shops, and journal special issues organization on subjects related to algorithms, opti-

mization, services for e-commerce, and cloud computing. Knowledge acquired by dif-

ferent postdocs working on the project will be also very valuable when looking for

academic positions or key industrial positions. This will also operate sibling and dis-

semination of knowledge.

We are convinced that by combining the experience of the groups and the con-

vergence of interests for research, design, and implementation, this project will make

major advances in the areas of operation research, optimization, cloud computing, and

e-business. Among the results expected in this and future collaborations, the following

can be also emphasized:

– Training of high-level researchers who can support the postgraduate computer sci-

ence program in Poland and Luxembourg.

– Establishment of favourable conditions for future projects that consolidate the Poland-

Luxembourg relation in the areas of Operational Research and Computer Science.

– Accumulating resources, expertise and previous scientiﬁc results of the two teams

to provide solutions to todays e-commerce (internet shopping) and cloud computing

(cloud brokering) problems.

– Publishing articles in highly recognized and prestigious international journals and

conferences.

IShOP will achieve the major goal of the POLLUX program that is covered by

fostering scientiﬁc quality in priority domains such as business service design, by op-

timizing business services strategies related to Internet shopping, cloud brokering, and

e-commerce websites. Internet shopping, which can be considered a sub-category of e-

commerce as it predominantly refers to business-to-consumer (B2C) transactions such

as online retail or online auctions, is one of the key business activities offered over the

Internet. At a time of economic crisis when consumers and businesses are under tight

ﬁnancial constraints, there is a shift towards e-commerce, and it is expected to have

more and more importance in the world economy in the coming years.

The IShOP consortium strongly believes in dissemination of research ﬁndings through

journal and conference publications. They will occur as a result of the training of Mas-

ters, PhD, and postdoctoral researchers. The IShOP consortium is aware of the growing

need to establish workable relationships with leading researchers. The workshop organi-

zation will help to involve highly recognized scientiﬁc researchers from the community

worldwide.

The consortium aims to achieve high impact and excellence well beyond the aca-

demic community by collaborating with external partners such as industry partners,

government bodies, institutions and international actors. The intended result of this

project is to seek cooperation with a business that could use elaborated ISOP results. At

the current state of art this is not possible, as models are too simpliﬁed, and they exclude

too many real life situations. The improvements are meant to meet the requirements of

large online shops (like Amazon) or price comparison sites (like Google Shopping,

Ciao!, Ceneo, etc.). Industrial session organization will strongly support relations with

external partners, as has been demonstrated by the success of Business Meets Research

forum in Luxembourg, to show the expertise, capabilities and innovative solutions of

the IShOP consortium.

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

Cloud computing is positioning itself as a new emerging platform for delivering

information infrastructures and resources as IT services on a pay-as-you-go basis to

streamline business processes. It has evoked a high degree of interest internationally,

with many challenges such as security, privacy and resource optimization remaining

open. Luxembourg is one of the best data center parks in Europe with more than 15

data centers and more than 35,000 m2 for IT rooms. The country is signiﬁcantly in-

creasing its level of cloud computing expertise and extending this skill Europe-wide.

This knowledge acts as a springboard for the future worldwide developments of the

Luxembourg computing cloud. However, due to the vast diversity in the available cloud

services, it has become difﬁcult for the customers to decide whose services they should

use and what is the basis for their selection. The IShOP consortium will contribute to

alleviate this problem by developing models, algorithms and mechanisms to the cloud

brokering problem. Such models and algorithms can make a signiﬁcant impact and will

create healthy competition among cloud providers to satisfy their service level agree-

ment and meet customers quality of service requirements.

IShOP will serve as the basis for educational material at Master level (module in

the problem solving and optimization lectures at UL as well as algorithm design and

electronic commerce dedicated lectures at PUT) and PhD level (seminars for the SnT,

for the CSC research unit and ILIAS laboratory and the Doctoral School at PUT). The

project will also develop and improve the transfer of knowledge between Poland and

Luxembourg.

Acknowledgements

This project is supported by the FNR (Luxembourg) and NCBiR (Poland), through

IShOP project, INTER/POLLUX/13/6466384.

References

1. Hagel III, J.: Net gain: Expanding markets through virtual communities. Journal of Interac-

tive Marketing 13 (1999) 55–65

2. Timmers, P.: Business models for electronic markets. Electronic Markets 8 (1998) 3–8

3. Coopers, P.: Your access to european market. [online] (2009) http://www.pwc.com/lu/en/

invest-in-luxembourg/docs/pwc-publ-lux-where-else.pdf.

4. Blazewicz, J., Kovalyov, M.Y., Musial, J., Urbanski, A., Wojciechowski, A.: Internet shop-

ping optimization problem. Int. J. Appl. Math. Comput. Sci. 20 (2010) 385–390

5. Blazewicz, J., Bouvry, P., Kovalyov, M.Y., Musial, J.: Internet shopping with price sensitive

discounts. 4OR-Q J Oper Res 12 (2014) 35–48

6. Blazewicz, J., Bouvry, P., Kovalyov, M. Y., Musial, J.: Erratum to: Internet shopping with

price-sensitive discounts. 4OR-Q J Oper Res (2014) online ﬁrst.

7. Blazewicz, J., Cheriere, N., Dutot, P. F., Musial, J., Trystram, D.: Novel dual discounting

functions for the Internet shopping optimization problem: new algorithms. J Sched (2014)

online ﬁrst.

8. Wojciechowski, A., Musial, J.: Towards optimal multi-item shopping basket management:

Heuristic approach. In Meersman, R., Dillon, T., Herrero, P., eds.: On the Move to Mean-

ingful Internet Systems: OTM 2010 Workshops. Volume 6428 of Lecture Notes in Computer

Science. Springer Berlin / Heidelberg (2010) 349–357

Internet Shopping Optimization Project (IShOP)

9. Blazewicz, J., Musial, J.: E-Commerce Evaluation – Multi-Item Internet Shopping. Op-

timization and Heuristic Algorithms. In Hu, B., Morasch, K., Pickl, S., Siegle, M., eds.:

Operations Research Proceedings 2010. Operations Research Proceedings. Springer Berlin

Heidelberg (2011) 149–154

10. Revelle, C., Eiselt, H., Daskin, M.: A bibliography for some fundamental problem categories

in discrete location science. European Journal of Operational Research 184 (2008) 817–848

11. Krarup, J., Pisinger, D., Plastriab, F.: Discrete location problems with push-pull objectives.

Discrete Applied Mathematics 123 (2002) 363–378

12. Eiselt, H., Sandblom, C. L.: Decision analysis, location models, and scheduling problems.

Springer (2004)

13. Melo, M., Nickel, S., Saldanha-da Gama, F.: Facility location and supply chain management

- a review. European Journal of Operational Research 196 (2009) 401–412

14. Iyigun, C., Ben-Israel, A.: A generalized weiszfeld method for the multi-facility location

problem. Oper. Res. Lett. 38 (2010) 207–214

15. Garey, M., Johnson, D.: Computers and Intractability: A Guide to the Theory of NP-

Completeness. New York, Freeman (1979)

16. Blazewicz, J.: Zło

zono

c obliczeniowa problem

ow kombinatorycznych. Warszawa:

Wydawnictwa Naukowo-Techniczne (1988)

17. Musial, J.: Applications of Combinatorial Optimization for Online Shopping. NAKOM,

Poznan (2012)

18. Buyya, R., Broberg, J., Goscinski, A. M.: Cloud Computing Principles and Paradigms. Wiley

Publishing (2011)

19. Foster, I., Zhao, Y., Lu, S.: Cloud Computing and Grid Computing 360-Degree Compared.

In: Grid Computing Environments Workshop, 2008. IEEE (2008) 1–10

20. Zhang, Q., Cheng, L., Boutaba, R.: Cloud computing: state-of-the-art and research chal-

lenges. Journal of Internet Services and Applications 1 (2010) 7–18

21. Rimal, B., Choi, E., Lumb, I.: A taxonomy and survey of cloud computing systems. In: 5th

International Joint Conference on INC, IMS and IDC. (2009) 44–51

22. Tantar, A., Nguyen, A., Bouvry, P., Dorronsoro, B., Talbi, E.: Computational intelligence

for cloud management current trends and opportunities. In: IEEE Congress on Evolutionary

Computation (CEC), IEEE (2013) 1286–1293

23. Kliazovicz, D., Pecero, J., Tchernykh, A., Bouvry, P., Khan, S., Zomaya, A.: Ca-dag:

Communication-aware directed acyclic graphs for modeling cloud computing aplications.

In: IEEE 6th International Conference on Cloud Computing, IEEE-Cloud 2013. (2013)

24. Chaisiri, S., Lee, B., Niyato, D.: Robust cloud resource provisioning for cloud computing

environments. In: IEEE International Conference on Service-Oriented Computing and Ap-

plications (SOCA). (2010) 1–8

25. Castro, H., Villamizar, M., Sotelo, G., Diaz, C., Pecero, J., Bouvry, P.: Green ﬂexible oppor-

tunistic computing with task consolidation and virtualization. 16 (2013) 545–557

26. Guzek, M., Varrette, S., Plugaru, V., Pecero, J., Bouvry, P.: A Holistic Model of the Perfor-

mance and the Energy-Efﬁciency of Hypervisors in an HPC Environment. In Pierson, J.M.,

Da Costa, G., Dittmann, L., eds.: Energy Efﬁciency in Large Scale Distributed Systems.

Lecture Notes in Computer Science. Springer Berlin Heidelberg (2013) 133–152

27. Pal, R., Hui, P.: Economic models for cloud service markets: Pricing and Capacity planning.

Theoretical Computer Science 496 (2013) 113–124

28. Grozev, N., Buyya, R.: Inter-cloud architectures and application brokering: taxonomy and

survey. Software: Practice and Experience 44 (2014) 369–390

29. Spillner, J., Brito, A., Brasileiro, F., Schill, A.: A Highly-Virtualising Cloud Resource Bro-

ker. In: IEEE/ACM Fifth International Conference on Utility and Cloud Computing. IEEE

Computer Society (2012) 233–234

EPS Vienna 2014 2014 - European IST Projects - The Quest for Excellence Towards 2020

30. Nesmachnow, S., S., I., Dorronsoro, B., Talbi, E. G., Bouvry, P.: List scheduling heuristics

for virtual machine mapping in cloud systems. In: VI Latin American Symposium on High

Performance Computing (HPCLatAm). (2013) 1–12

31. Nesmachnow, S., S., I., Dorronsoro, B., Talbi, E. G., Bouvry, P.: A parallel hybrid evolution-

ary algorithm for the optimization of broker virtual machines sublet in cloud systems. In:

2nd International Workshop on Soft Computing Techniques in Cluster and Grid Computing

Systems (SCCG). (2013) 1–12

32. Usha, M., Akilandeswari, J., Fiaz, A. S.: An efﬁcient qos framework for cloud brokerage

services. In: International Symposium on Cloud and Services Computing (ISCOS). (2012)

76–79

33. Carpentier, J., Gelas, J., Lefevre, L., Morel, M., Mornard, O., Laisne, J.: CompatibleOne:

Designing an Energy Efﬁcient Open Source Cloud Broker. In: Second International Confer-

ence on Cloud and Green Computing (CGC). (2012) 199–205

34. Nesmachnow, S., Dorronsoro, B., Pecero, J., Bouvry, P.: Energy-Aware Scheduling on Mul-

ticore Heterogeneous Grid Computing Systems. Journal of Grid Computing 11 (2013) 653–

680

35. Iturriaga, S., Nesmachnow, S., Dorronsoro, B., Bouvry, P.: Energy efﬁcient scheduling in

heterogeneous systems with a parallel multiobjective local search. Computing and Informat-

ics Journal 32 (2013) 1001–1022

36. Pinel, F., Dorronsoro, B., Pecero, J., Bouvry, P., Khan, S.: A two-phase heuristic for the

energy-efﬁcient scheduling of independent tasks on computational grids. Cluster Computing

16 (2013) 421–433

37. Alba, E., Dorronsoro, B.: Cellular Genetic Algorithms. Volume 42 of Operations Re-

search/Computer Science Interfaces Series. Springer-Verlag US (2008)

38. Tantar, A. A., Danoy, G., Bouvry, P., Khan, S. U.: Energy-Efﬁcient Computing Using Agent-

Based Multi-objective Dynamic Optimization. In Kim, J.H., Lee, M.J., eds.: Green IT: Tech-

nologies and Applications. Springer Berlin Heidelberg (2011) 267–287

39. Dorronsoro, B., Danoy, G., Nebro, A. J., Bouvry, P.: Achieving super-linear performance in

parallel multi-objective evolutionary algorithms by means of cooperative coevolution. Com-

puters & Operations Research 40 (2013) 1552–1563

40. Nebro, A., Durillo, J., Luna, F., Dorronsoro, B., Alba, E.: MOCell: A Cellular Genetic

Algorithm for Multiobjective Optimization. International Journal of Intelligent System 24

(2009) 726–746

41. Nebro, A., Luna, F., Alba, E., Dorronsoro, B., Durillo, J., Beham, A.: AbYSS: Adapting

Scatter Search to Multiobjective Optimization. IEEE Transactions on Evolutionary Compu-

tation 12 (2008) 439–457

42. Blazewicz, J., Ecker, K. H., Pesch, E., Schmidt, G., Weglarz, J.: Handbook on Scheduling:

From Theory to Applications. International Handbooks on Information Systems. Springer

Science & Business Media (2007)

43. Pinel, F., Dorronsoro, B., Bouvry, P.: Solving very large instances of the scheduling of

independent tasks problem on the gpu. Journal of Parallel and Distributed Computing 73

(2013) 101–110

44. Pecero, J., Huacuja, H., Bouvry, P., Pineda, A., Loces, M., Barbosa, J.: On the energy op-

timization for precedence constrained applications using local search algorithms. In: In-

ternational Conference on High Performance Computing and Simulation (HPCS). (2012)

133–139

Internet Shopping Optimization Project (IShOP)